Learning Tensorflow

Dhanya Arumughom -

TensorFlow is an open source software library from Google widely used for designing deep neural network models and large scale machine learning models. It was developed by the Google Brain team for use within Google’s Machine Intelligence research organization and was called DistBelief then. The system is general enough to be used for a wide variety of domains and Google open sourced it in 2015 making it available to everyone for building deep learning applications. It is written in C++ , Python and CUDA.

What exactly is TensorFlow all about?

TensorFlow is an ecosystem or probably a full stack with entire suite for developing deep learning models. Deep learning models deals with machine learning based on learning data representations. Deep learning models in fact mimic the working of human brain in processing data and analyze the patterns to arrive at decisions. Probably you know this; the neural networks in brain are responsible for the pattern detection by humans. Deep learning methods use hierarchical level of artificial neural networks (ANN) commonly referred to as deep neural networks as it is comprised of multiple layers.

A deep learning application output is the result of a tuned network created by optimization processes involving repeated computation of complex mathematical operations. TensorFlow approaches this series of computations using a data flow graph which considers the computations as a flow of data. Each node in the graph is the mathematical operation and edges of the graph represent multi-dimensional data sets on which the operations are performed. A multi dimensional array of data or matrices is called a ‘Tensor’ and hence the name TensorFlow.

Tensor flow is a designed to be a low level library for computations which can perform addition, matrix multiplication, etc to implement an algorithm and allows users to build various models from scratch. This is in contrast to machine learning libraries like Scikit-learn which are high level libraries with machine learning algorithms already available in it making it not that useful for more complex Deep learning problems.

Building the data flow graph is the key ingredient in TensorFlow. The TensorFlow flow includes building the computational graph, Initializing variables, creating sessions and running the graph in the session. A placeholder allows you to create or build the computation graph without the data and values for the nodes can be fed in at the execution time. TensorFlow provides APIs in several languages to create and execute the TensorFlow graphs, like Python, C++, Java and Go. But the Python API is the most complete and stable one as of now.

The TensorFlow ecosystem consists of three modules:

  • TensorFlow (API): Contains the APIs to define deep learning models and train the models with the data sets. It has High level and Low level APIs.

  • TensorFlow Serving: A flexible and high performance model server system for production environments. This module makes it easy to deploy new algorithms, while keeping the same server architecture and APIs and thus making switching from old models to new models easy without any downtime.

  • TensorBoard: A solution to visualize the TensorFlow graph. It helps in analyzing and debugging the graphs.

Why TensorFlow?

With many other deep learning libraries around like the Theano and Torch, what makes TensorFlow unique? Tensor flow is a pretty new tool which is still under development for machine learning/deep learning, but has gained the attention of a lot of engineers very quickly. Let’s take a quick glance on some of the major advantages of TensorFlow over other deep learning libraries.

One of the main reasons for the popularity is that it is backed by Google which has proved its excellence in areas of Deep learning and AI. Imagine the depth of Artificial intelligence in the Google photos app, Google maps, Google translate or the search experience provided by Google. TensorFlow is used in all these recognized products of Google.

One feature that is claimed to be the highlight of TensorFlow and is not available in most of the other libraries like Theano or Torch is the distributed training ability. TensorFlow has the ability to perform partial sub graph computation which allows the partition of a neural network so that distributed training is possible. Thus TensorFlow allows you to deploy the computation to CPUs or GPUs easily for distributed learning.

Another important attraction of Tensor flow is the flexibility offered by the TensorFlow graph architecture using low level API. You can literally create any machine learning model using simple operations which are like the basic building blocks of your system making TensorFlow highly extensible. TensorFlow is also highly portable and runs on CPUs, GPUs, desktops, server and mobile computing platforms. TensorFlow lite is a software pack specifically designed for mobile and embedded devices.

Thus, TensorFlow turns out to be a very powerful and high performance library for Deep Learning and numerical computations for creating trained models. TensorFlow is gaining momentum for various applications in voice recognition, Image recognition, video detection, text based applications like language detection and text summarization, etc.

Learn TensorFlow

The areas of machine learning and artificial intelligence are evolving at a high pace and the more advanced deep learning algorithms have become an area of interest lately. TensorFlow has become one of the primary tools for deep learning with many of its benefits, ease of use and the support from a strong community.

It is definitely a tool that can help you build a profile and take you to new heights in deep learning field. Learning TensorFlow requires you to have a good grasp of Machine learning, deep learning and artificial neural networks concepts. And of course, you should have good knowledge of statistics, algebra and mathematics for learning machine learning and ANN concepts.

Also, you have to be comfortable with Python too as it is the recommended API for TensorFlow as now even though it offers other language options.

Best Tutorials For Tensorflow

We have collected a list of some of the best beginner level tutorials on Tensor Flow.

Best Courses For Tensorflow

However, if you already decided to take a deep dive, here are some of the best courses we found on Tensor Flow.

For Beginners
For Intermediate Level
For Advanced Level